Ultra Precision Reference LTZ1000

[CalMachine]

Keysight 34470A Multimeter Teardown by Dave Jones, on Flickr

[AG7CK]

It has been mentioned several places on the forum: It is the autocal reference resistor.

[TiN]

Probably comes from LTFLU-based KI 7510 price?

[MisterDiodes]

This is getting off topic now, but in answer to your question:

In the US, it's more like a little over $4k per meter when you get the warranty, shipping, sales tax (depending on what state) and TruVolt software stuff turned on.  Z540 cal goes on top of that.  The lower price doesn't include the advanced software license - at least not when we got the sales demo and the written quote from Keysight.

Note the "Typical Configured" price of over $3800:
http://www.keysight.com/en/pd-2520154-pn-34470A/digital-multimeter-7-digit-performance-truevolt-dmm?cc=US&lc=eng

And again - at least when we patiently waded thru the sales demo in 2015, this meter -was not- nearly as stable in 6.5 digit mode as an LM399-based 3456a, which you can get for maybe around $300 (or somewhere around that) on eBay (a basic 3456a cal runs about $100 or $175 for Z540).  In fact the demo 34470a not only drifted faster in four hours as did a 3456a drifts in a month (new '56a spec was 8ppm/day, older ones we have all measure at < 4ppm per month and are 1ppm or less per day), the 34470a drifted faster in four hours than all FIVE 3456a's 30 day spec in the same room.  And of course against the 3458a's there is no contest, but the 34470a is not in competition with that one.  Tell me again why you'd want this meter if you're after good stability over even 24hrs - and why is it a an example of a good Vref module design technique?  At any price??

Here's a real sample in the lab:  In comparison to a used, working Z540 calibrated 3458a that we've got a total of about $4300 invested after first year (24hr hr spec of 0.5ppm plus .05ppm of range on 10VDC range - which is what about all quality LTZ -based devices are capable of, with or without voodoo slots on the Vref module)...and on 34470a you get 8ppm + 2ppm (10DCV range) per 24 hrs (see datasheet):

http://www.keysight.com/en/pd-2520154-pn-34470A/digital-multimeter-7-digit-performance-truevolt-dmm?cc=US&lc=eng

Also note that 34470a input bias current is "<30pA"...So we'll take that as meaning up to 29.9pA, which means you're not going to be throwing out your decent analog null meter anytime soon (Keithley 155's / Fluke 845's are < 200fA and even well below that if they are dialed in carefully).

So the disappointing question I'm left with is this:  If you're Keysight, and you have all the design and development work done on a perfectly good working 3458a LTZ module where you already have lots of inventory - and at this point your Vref cost is really only the BOM parts cost - why on Earth do you try to design a medicore, cheap-ass LTZ Vref to build a 7.5 digit meter with only medicore stability?  There wasn't say $10~$15 (volume cost for better resistors and caps) left over in the budget to use a PROVEN GOOD LTZ module design you already own that requires no engineering dev work?  Software will never compensate for higher drift rates on an LTZ module - In other words your LTZ-based device is never going to be more stable than the Vref module itself.

Maybe the rest of the meter is so dismal that they can't get better than even 2 or 4ppm per day drift (which is still not that great). Maybe they just gave up trying at some point.

Again - people might not need the high stability and they like the smaller size / faster reading per second of 34470a.  And that's what they should get.  The 34470a serves as a good example of "Build to a price point" LTZ module technique that gets you relative medicore stability performance.  You get what you pay for, and sometimes if you pay less for an older, much more stable unit - you're getting more.  IF real stability and good quality binding posts are more important to you than the fluffy display - then 34470a is not a really a good choice for a lot of low-ppm measures.

[VK5RC]

I just had a 'naughty' thought, perhaps they did try the 3458's LTZ module in the 34470 and perhaps it worked a bit too well!

[Edwin G. Pettis]

Possibly, they had to use a LTZ circuit, cheapened as it is, because a LM399's noise was probably just a little too much even for selected units and they were trying to hit a 'usable' 7.5 digit claim, even though the specs don't support real 7.5 digit performance.

[Kleinstein]

Not all cheap resistor are settling to a stable value - some continue drifting in one direction (like up with most carbon resistors). Even with relatively poor resistors the LTZ1000 reference is relatively stable.

So TC of the 34470 is likely more from the rest of the circuit. This could be resistors or not perfectly matching FET on resistance in a few places. The reading coming back with an ACAL confirms the reference itself it not the problem. It would be only after ACAL that you get mainly drift of the reference circuit, and very little from the rest of the meter.

[floobydust]

The Keysight plastic carrier does not do anything for LTZ1000 heat loss due to conduction on the FR-4 of the reference PCB, which has variable airflow directly across it?
I don't know the 3458a mechanical but the ref pcb seems out of direct airflow, in the corner. I see a duct to ensure that.

Let's cool the heater.

[Kleinstein]

I won't expect the temperature control loop to get unstable (in the sense of start oscillation) with too much isolation. It is more like running out of regulation because self heating from the zener current. So a higher set point temperature would be needed. With the resistor as heater and thus a square law heater, the thermal loop gain will go down with lower power. So the loop will get more sluggish and less accurate at lower power / more isolation.

Besides the thermocouple effects at the Kovar pins, also the TC of the pin resistance could be a factor. Here some extra thermal isolation on the board would help, but is limited due to the copper connections.

[floobydust]

LTZ1000ACH ship dates, Digi-Key keeps pushing them back from Sept. to mid November now. LT store says "consult factory". Which one of you is hoarding them.

I used to gripe about LT prices being a bit higher and their field engineers told me they are a smaller fab and always have stock, it's part of the cost.
To me it is important to have availability for sole-sourced parts, so I understood.

But not today, perhaps these parts have a long burn-in or are on allocation now.
I truly hope AD didn't mess things up.

[kj7e]

LTZ1000ACH ship dates, Digi-Key keeps pushing them back from Sept. to mid November now. LT store says "consult factory". Which one of you is hoarding them.

I used to gripe about LT prices being a bit higher and their field engineers told me they are a smaller fab and always have stock, it's part of the cost.
To me it is important to have availability for sole-sourced parts, so I understood.

But not today, perhaps these parts have a long burn-in or are on allocation now.
I truly hope AD didn't mess things up.

Ive had one on back-order from Digikey for about 2 months.  The ETA has not changed and is still said to ship later next week.  So I suspect they simply ran out, then back-orders depleted their next stock shipment.  Digikey does not seem to stock many of these, never seen their stock more than about 70.  And I'm sure Digikey and other large company's will get first dibs on any stock from LT before the direct on-line store.

[branadic]

No problem with the non-A version, they stock 106. So why you want to go for the A version?

-branadic-

[floobydust]

I see a lot of struggle with thermal design alone, for the LTZ1000. An 84 page thread with 571,940 views and I didn't see a thermal model or mechanical configuration solidified.

It's difficult to do thermal design when conduction, convection, and radiation heat losses are unknown.
The LTZ1000 datasheet offers one number, ?JA for us to work with.
Graphs G04,G05,G06 show about 30-40% heater power required for the "A" compared to non-A version (but no mention of can configuration; on a PCB, free air, cap on top etc.)

I think the A's higher thermal resistance 400°C/W vs 80°C/W as a "Matryoshka doll" approach insulating the die from the case greatly lowers heater power requirements which might lower the Kovar TC effects due to a smaller temperature gradient. And the cap/insulation delta T. I'm not sure, it's my speculation.
Ultimately I think the thermal design is more forgiving for the A part.

[martinr33]

Part of the problem is that the stability goal of these designs is such that it may be difficult to simulate the system. In particular, the PCB traces may have quite an influence.

The voltage temperature coefficient for copper to Kovar is 40 microvolts per degree Celsius. As we are looking for sub-microvolt stability, and assuming that the temperature at the can end is consistent, that means we want the temperature of the critical pads to vary by less than .01 C.

That said, top and bottom caps seem to do a lot of the work. Plus, getting some separation from other components will help. Also suggests that flush mount is best, as it will reduce the temperature difference across the leads. Second up would be making sure that the sensitive pins have a similar thermal path - which suggests thin traces connected to the critical pins. There's also the issue of the external resistors, which may be quite a lot more sensitive to temperature shifts.

[floobydust]

I agree, top and bottom covers are beneficial but I wonder about the reference IC-PCB interface.

Talking about one temperature gradient- between the Kovar leads and the PCB.
Silicon-gold-Kovar-solder-ENIG-copper; Don't know if this gives the usual quoted 35-40?uV/°C and are people using low EMF solder here.

For the Agilent 34470a implementation (above pics) I think it's not optimal to insulate the LTZ case but have the PCB exposed to forced air, as the FR-4 acting as a heatsink. Voodoo slots would lower conduction heat losses from the LTZ leads, but not evenly across all pins.

My analogy is standing bare foot in cold water, and you have a winter jacket on.
Your core is nice and warm but a large temperature gradient down to your feet.

I think you need to enclose (insulate) the LTZ (on both PCB sides) and as much of the ref PCB around it as possible to get isothermal IC connections. So the temp gradient occurs away mostly on the PCB copper traces leading to the ref IC pads. Or maybe not.
Part of the fun here is learning a new microcosm. I think I will add the connecting TC's and try Spice sims when I get a chance.

[martinr33]

The analogy is slightly different - we're talking about your feet being in separate buckets of cold water. If they are not at the same temperature, you start getting the 40uV per degree C of the Kovar working against you. Everything else in the path is either at the same temperature (gold wires in the can) or relatively thin such that the temperature delta is small (the kovar - solder - copper layers at the PCB interface).

So our challenge is, getting the critical PCB pads at the same temperature.

[floobydust]

I understand your point. This thermal IR image from TiN's KX Ref at xdev.com shows the main (~12°C) temperature gradient from top (IC case) to bottom (PCB). I wrongly thought that was something to minimize as I realized LT's intent to cancel out thermal EMF's, the Seebeck pairs:
pin 3, 7 cancel
pin 4, 5 cancel
pin 7, 6 cancel
pin 6, 8 cancel

Between pins it would be voodoo slots and differing copper traces making pads unequal temperature. That would cause large errors if I'm understanding this correctly now.

[martinr33]

That's it exactly.

If the pads are reasonably well insulated from the rest of the board and have good thermal conductivity between them, then short legs should work best. That approach assumes that the pads have a materially higher temperature than the rest of the system.

If the pads are not well insulated, then longer legs will help keep the temperature more even at the kovar-PCB junction. TiN's picture seems to show a degree or two caused by the package, even with quite long Kovar legs.

The Datron approach seems to be short legs; a big insulator on the back; and slots. A ground plane inside the can pinout, populated with through-plated holes, would increase the thermal conductivity between the pins. I'd bet that, with a good insulator on the back, this would be better than slots. When these things were being designed, that was not an option.

Kleinstein's copper cap is less thermally efficient, but should help with even distribution of temperature.

And the big nono is unbalanced copper coming out of the pads.

As noted elsewhere, the 3458a reference works well. You see there long, thin traces and insulating caps that cover both sides of the board. The LTZ1000 also sits well clear of any other device, and the separate board does not hurt.

The 34470A design appears to implement some of these ideas.
https://youtu.be/_QpApuKdcqQ?t=887

[floobydust]

Theoretically, 0.36°C temperature difference between two LTZ pads (pins 3,7) Kovar generates 14uV which is 2ppm error, me thinks.
If this is correct, then some PCB designs out there are aggravating this by not having symmetrical copper traces leading away from the pads. The payoff from guard bands, copper pours, star-connections etc may get lost.


Looking at Keysight 34470-26503 REV 003 ref. PCB; Along with others, I'm not impressed...

Underside has extra "pads" at LTZ pins 3, 7; I think the idea is to thermally balance those pads.
You'd not put a bypass cap there- no solder mask and detrimental to long term drift as discussed earlier in this thread.

But star connection on pin 3 but not pin 7. Strange. Three ferrite beads for EMI but skinny traces to GND.

I sketched a partial schematic and could not make sense of it. Labels on the backside make it difficult.
U1 LT1013 pin 5 and pin 2 have RC between them? Must be going crazy, that's between the heater and ref op-amp. {used wrong LT1013 pinout SOIC vs DIP}


{pic is crop of Dave's teardown pic.}

[chuckb]

Keep in mind that the SMD version of the LT1013 has a non standard pin out. Pin 5 is an output and pin 2 is the negative supply for the chip.  They had to do this to fit the die inside a standard SMD package. It has caused me no end of problems...

[martinr33]

The mixed-up design ideas is something you see when the design engineer and the layout engineer are not talking. For a 7.5 digit DMM, 2uV is only 2 counts. There's likely more noise than that in the rest of the system. If there's a new 8.5 digit coming, then maybe some of this gets fixed. On the other hand, the cover area is quite large and may be enough to keep the temperatures even despite the thermal path.

[floobydust]

I did have the wrong LT1013 pinout (DIP vs SOIC); Now 34470-26503 schematic looks vanilla except for the FB's.
I'm surprised at the cost reduction with what looks like Yageo 0603's and three precision 1206 's. Not sure who uses Helvetica font on their 1206's. Stackpole maybe.
If Agilent broke the piggy bank, using three Vishay VFCP smt foil adds ~$30 cost.

[Kleinstein]

A temperature difference between the pins is not good, but it only the change in the temperature difference that would cause trouble. The constant part would be just an other small addition to the reference voltage (even with very low noise).

One can expect the temperature differences to be about proportional to the heater power and thus about proportional to the temperature difference of the reference to environment. If the assumed 0.36 K difference is at a ref to air difference of 20 K, the 2 ppm effect would occur over a 20 K span. Thus it would be a contribution to the TC of 0.1 ppm/K.  HP tends to use a higher temperature for the LTZ1000 - thus the effect would be even smaller.  So the additional error is in the range of the normal TC.

If made on purpose and well sized it could even be a good alternative to the kind of afterthought 400 K resistor to compensate residual TC of the non A version. With a FEM simulation (or a few iterative steps or pure luck) it sounds plausible to find such a suitable layout / thermal design.

The rather cheap SMT resistors might be the bigger issue, e.g. due to humidity effects on the board.

[Dr. Frank]

Hi floobydust,
would you mind publishing your schematic here?

Thanks
Frank

[floobydust]

Here is my schematic of the Keysight 34470-26503 REV 003 reference PCB, from Dave's teardown pics.

Connector J1 pinout I could not see traces and the labels are in the way. Surely some errors there. If people let me know I will update it.


Comparing to HP3458a reference, I notice:

Collector resistors 74k25 -> 75k3 (look like vanilla 1%'ers)
Zener resistor 111R-> 100R
GND balance resistor 2k76-> 4k22

Heater base drive 499R-> 511R
Heater ratio still 15k/1k
Heater monitor V at transistor instead of op-amp
Using dual diode BAS28
No temp slope comp resistor for non-A LTZ but support for TO-126 heater transistor
Addition of three ferrite chip beads

UPDATE 2020-12-26: schematic corrections and updated to Rev. 2 based on info from thread about 34470a LTZ1000 reference PC board clones:
https://www.eevblog.com/forum/metrology/buyers-beware-misleading-34470a-reference-modules-(clones)-on-ebay/msg3383376/#msg3383376